August 14, 2018

Status of AI and Machine Learning in Radiology

Commentary
Tom Parsons

News and hype surround the field of radiology with headlines around the world purporting that it will be disrupted overnight. Few companies though really have the evidence to back up these claims.

A combination of factors have led to this field being a target for innovators including the expansion of image archiving, the increase of diagnostic image-sharing and the computer-readable DICOM format.

These innovative companies are seeking to apply AI, Machine and Deep Learning to this field in the hope of achieving time and cost savings, and to help doctors detect changes such as tumors, hardening of the arteries and provide highly accurate measurements of organs and blood flow.

Even though in principal the challenges in this field are ripe for the application of modern technology, there are considerable market barriers new companies must face. These barriers include regulatory challenges, how the technology is integrated into the clinical workflow and the learning curve for those clinicians who must be trained to use the tool.

These barriers have not held some innovative companies back, below we highlight some of the best of these technologies as well as hear from experts in the field on what to expect in the future.

Selection of Recent Regulatory Winners

Arterys, is developing imaging analytics software solutions deployed via their cloud-based medical imaging deep learning platform. Arterys recently received FDA Clearance for a broad Oncology Imaging Suite which covers all solid tumors, its fifth 510(k) clearance from the FDA. The first deep learning workflows will be for lung CT scans, and liver MRI and CT scans. Using this solution, radiologists can easily confirm, evaluate, quantify, and report on the absence or presence of lung nodules and liver lesions along with their key characteristics using a simple web browser. GE Ventures is an investor.

“The potential of AI is not only to improve clinical insights and productivity for both the radiologist and the provider facility, but it also extends to expanding geographic access to advanced diagnostic capabilities and service lines so that more patients can benefit. The market is early in the adoption of these solutions, and their full impact is only beginning to be understood and appreciated.” - Leslie Bottorff, Managing Director, GE Ventures

“The potential of AI is not only to improve clinical insights and productivity for both the radiologist and the provider facility, but it also extends to expanding geographic access to advanced diagnostic capabilities and service lines so that more patients can benefit. The market is early in the adoption of these solutions, and their full impact is only beginning to be understood and appreciated.”

Zebra Medical Vision is developing algorithms to assist radiologists in detecting often overlooked indications in CT scans. Their algorithms can be applied to detect low bone mineral density, breast cancer, fatty liver, coronary artery calcium, emphysema with a flat rate of $1 per scan analysed in their cloud. This week Zebra announced the CE regulatory approval of its newest algorithm to be included in its growing Deep Learning Imaging Analytics platform. The algorithm, capable of detecting Intracranial Hemorrhages is the latest addition to their “All-In-One” AI1 business model.

It’s All About Workflow

Getting an algorithm certified is just the first step. It then has to be integrated into the providers existing systems. Ideally, AI will run automatically, producing discrete assessments that the radiologist can validate. Once approved, the data can be pulled into the EHR, where downstream providers can act accordingly. Several of the imaging analytics start-ups including Arterys and Zebra have developed cloud-based platforms for integration into the existing workflow. However, even with cloud infrastructure, there may be challenges associated with integrating these AI platforms with providors systems.

"When grasping the whole spectrum of radiomics tools and applications, it will be challenging to seamlessly integrate into radiology workflow" - Faiq Shaikh M.D, Director Life Sciences Research,Clusterone, Inc.

“Its an incredibly exciting time for AI applications, particularly in imaging…..However, I think what people need to understand is that there are already very large numbers of algorithms that are out there and have been developed over the last 20-30 years or so. What we really need is a better delivery mechanism to be able to create a platform so that they can be disseminated and actually available and implemented in daily practice. What one really needs is the ability to index those files, navigate databases, to be able to task and develop those algorithms for a particular area” Eliot Siegel M.D, Professor of Diagnostic Radiology and Nuclear Medicine, University of Maryland

Clinician Engagement is Key

Initially media on AI in radiology has conjured up fear of computers replacing radiologists. The focus now, is less on how to avoid it, and more on how radiologists can leverage it successfully.

“Most projects and startups focus on big data in imaging analytics which is the obvious perspective. I think, however, an even more interesting perspective is a patient's serial data especially in cardiology with echocardiograms. What the clinicians need is an analytical methodology in which any changes in echocardiogram month to month/ year to year can be easily delineated. Rarely is this being addressed and this is yet another example of a lack of clinician-data scientist synergy." - Anthony C. Chang M.D, Chief Intelligence and Innovation Officer, Children’s Hospital of Orange County

Companies have made great progress with bringing AI imaging analytics solutions to market, however in many cases it remains to be seen if these solutions will perform as expected in real-world settings. Delivery and infrastructure mechanisms need to be optimized as they are critical for clinical adoption and success in this space.